US20080027888A1

US20080027888A1 - Optimization of fact extraction using a multi-stage approach

Info

Publication number: US20080027888A1
Application number: US11/496,650
Authority: US
Inventors: Saliha Azzam; Kevin William Humphreys
Original assignee: Microsoft Corp
Current assignee: Microsoft Technology Licensing LLC
Priority date: 2006-07-31
Filing date: 2006-07-31
Publication date: 2008-01-31
Anticipated expiration: 2026-07-31
Also published as: EP2050019A1; TWI431493B; JP5202524B2; MX2009000588A; NO20085387L; JP2009545808A; RU2009103145A; BRPI0714311A2; RU2451999C2; AU2007281638A1; TW200817947A; AU2007281638B2; WO2008016491A1; US7668791B2; EP2050019A4

Abstract

Facts are extracted from electronic documents by recognizing factual descriptions using a fact-word table to match to words of the electronic documents. The words of those factual descriptions may be tagged with the appropriate part of speech. More detailed analysis is then performed on those factual descriptions, rather than on the entire electronic document, and particularly to the text in the neighborhood of the fact-word matches. The analysis may involve identifying the linguistic constituents of each phrase and determining the role as either subject or object. Exclusion rules may be applied to eliminate those phrases unlikely to be part of facts, the exclusion rules being based in part on the linguistic constituents. Scoring rules may be applied to remaining phrases, and for those phrases having a score in excess of a threshold, the corresponding sentence part, whole sentence, paragraph, or other document portion may be presented as representing one or more facts.

Description

BACKGROUND

Electronic documents may contain a mixture of facts and opinions. At times, a reader may only be interested in facts, or may wish to have the facts be identified. For example, a user performing an on-line search for information may wish to obtain facts about a particular subject as quickly and efficiently as possible. However, presenting a list of web pages or other electronic documents that are related to the search terms used require the user to individually examine each web page or other electronic document and distinguish the facts from the opinions or subjective information.
Attempts have been made to perform fact extraction. However, accurate fact extraction can be a slow and inefficient process even for high-speed server computers. Such fact extraction attempts generally apply a linguistic analysis to the entire contents of the electronic document to extract those facts that it may contain. When applying fact extraction to hundreds or thousands of electronic documents, the amount of time needed to achieve a result may be unacceptable.

SUMMARY

Embodiments provide optimization of fact extraction by using a multi-stage approach. The electronic documents are scanned to find factual descriptions that are likely to contain facts by using a fact-word table to match terms within sentences of the electronic documents to obtain a set of factual descriptions. Further analysis may then be performed, including determining linguistic constituents, e.g., syntactic constituents and/or semantics, in the neighborhood of that set of factual descriptions rather than on the entire document. Accordingly, time is saved by avoiding a complex lexical and syntactic analysis of the entire document for every electronic document of interest.
This Summary is provided to introduce a selection of concepts in a simplified form that are further described below in the Detailed Description. This Summary is not intended to identify key features or essential features of the claimed subject matter, nor is it intended to be used as an aid in determining the scope of the claimed subject matter.

DESCRIPTION OF THE DRAWINGS

FIG. 1 shows an example of a computer system for implementing embodiments.

FIG. 2 shows an example of an operational flow of a search involving the presentation of facts that have been extracted prior to the search.

FIG. 3 shows an example of an operational flow of a search involving the presentation of facts that have been extracted during the search.

FIG. 4 shows an example of an operational flow of the multiple steps of fact extraction.

FIG. 5 shows an example of a more detailed operational flow of the multiple steps of fact extraction.

FIG. 6 shows an example of a screen display providing search results that include the presentation of facts obtained from electronic documents discovered by the search.

DETAILED DESCRIPTION

Embodiments provide for fact extraction using multiple stages to avoid performing complex analyses of the entire documents of interest. Factual descriptions of the documents are recognized in relation to a fact-word table in an initial stage. These factual descriptions may be tagged with their parts of speech, either noun or verb. Then more detailed analyses may be done in a subsequent stage over those factual descriptions to thereby avoid such detailed analyses over the entire documents of interest. The linguistic constituents for each factual description may be determined and then exclusions and scores may be used to eliminate factual descriptions that are less likely to be facts. The factual descriptions remaining after the exclusions and scoring may then be presented as fact.
FIG. 1 shows an example of a computer system 100 that provides an operating environment for the embodiments. The computer system 100 as shown may be a standard, general-purpose programmable computer system 100 including a processor 102 as well as various components including mass storage 112, memory 104, a display adapter 108, and one or more input devices 110 such as a keyboard, keypad, mouse, and the like. The processor 102 communicates with each of the components through a data signaling bus 106. The computer system 100 may also include a network interface 124, such as a wired or wireless connection, that allows the computer system 100 to communicate with other computer systems via data networks. The computer system 100 may alternatively be a hard-wired, application specific device that implements one or more of the embodiments.
In the example, of FIG. 1, the processor 102 implements instructions stored in the mass storage 112 in the form of an operating system 114. The operating system 114 of this example provides a foundation upon which various applications may be implemented to utilize the components of the computer system 100. The computer system 100 may implement a search engine 118 or similar application for finding electronic documents relevant to a particular situation. For example, the search engine 118 may receive search terms entered directly through input device 110 by a user of the computer system 100 or may receive search terms submitted by a user of a remote computer that are received via the network interface 122.
The search and/or fact extraction may occur in relation to one or more sets of electronic documents that contain textual information such as web pages, standard word processing documents, spreadsheets, and so forth. These electronic documents may be stored locally as electronic document set 116. These electronic documents may also be stored at a non-local location such as network-based storage 124 containing an electronic document set 126. Network-based storage 124 is representative of local network storage, on-line storage locations of the Internet, and so forth. The network-based storage 124 is accessible via the network interface 122.
Additionally, these embodiments provide logic for implementation by the processor 102 in order to extract the facts from the electronic documents 116, 126. A fact extraction tool 120 may be present on the local storage device 112, either as a component of the operating system 114, a component of the search engine 118 or other application, or as a stand-alone application capable of producing its own independent results. The logical operations performed by embodiments of the fact extraction tool 120 are discussed below in relation to FIGS. 2-5.
The computer system 100 of FIG. 1 may include a variety of computer readable media. Such computer readable media contains the instructions for operation of the computer system and for implementation of the embodiments discussed herein. Computer readable media can be any available media that can be accessed by computer 100 and includes both volatile and nonvolatile media, removable and non-removable media. By way of example, and not limitation, computer readable media may comprise computer storage media and communication media.
Computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by computer system 100.
Communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. The term “modulated data signal” means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Combinations of the any of the above should also be included within the scope of computer readable media.
FIG. 2 shows an example of logical operations performed by a search engine 118 in conjunction with the fact extraction tool 120. In this example, the fact extraction tool 120 is utilized prior to a search occurring in order to generate a library of facts present in the electronic documents to be searched. In this manner, there is no processing time required to extract the facts but instead those facts have already been extracted and are retrieved from a fact library on the basis of the search terms entered.
The logical operations begin at collection operation 202 where the collection of electronic documents is obtained or access is otherwise achieved. For example, the electronic documents to eventually be searched may be saved to local storage or may be acquired via on-line access. The fact extraction tool 120 then operates upon each one of those electronic documents to attempt to extract all of the facts that are present in the electronic documents. The fact extraction tool 120 may generate a library of facts that are stored in association with the corresponding electronic documents and are available for access during future searches. For example, Table 1 shows such a library of associations.

	TABLE 1

	Electronic Document	Facts

	www.sample1.com	Fact A
		Fact B
		Fact C
	www.sample2.com	Fact AA
		Fact BB
		Fact CC
	www.sample3.com	Fact AAA

Continuing with the operational flow of FIG. 2, a user wishing to do a search to find relevant electronic documents, and particularly to find relevant facts from those electronic documents, enters a search term into the search engine 118 at term operation 206. In this example, the search engine 118 then searches through the electronic documents for the search terms and finds matching documents at document operation 208. The search engine also finds the previously extracted facts which match the search terms from those matching electronic documents and then displays the relevant documents or a link thereto along with the relevant facts at display operation 210. For example, a search term may be found in www.sample1.com and the search term may also be found to match Fact A and Fact B such that a link to www.sample1.com is displayed along with Fact A and Fact B. Thus, the user is quickly provided with facts related to the search terms that were entered. An example of such a screen display is discussed below in relation to FIG. 6.
Of course, as an alternative the search may be for previously extracted facts only, rather than for the electronic documents themselves. Furthermore, in certain circumstances the previously extracted facts may match the search terms regardless of whether the electronic documents containing the facts match the search terms.
FIG. 3 shows another example of logical operations performed by a search engine 118 in conjunction with the fact extraction tool 120. In this example, the fact extraction tool 120 is utilized during a search in order to discover facts present in the electronic documents as they are being found by the search. In this manner, there is no need for pre-search fact extraction and no need for storage of a library of facts. In such a scenario, the fact extraction tool may only scan snippets or summaries of the document to provide very fast results, or the entire document may also be scanned to extract all potential facts.
The logical operations begin at term operation 302 where a user enters a search term into the search engine 118. In this example, the search engine 118 then searches through the electronic documents for the search terms and finds matching documents at document operation 304. The extraction tool 120 is then employed at extraction operation 306 in order to analyze the electronic documents that have been found by the search in order to extract facts from those documents that are relevant to the search terms. The result of extraction operation 306 may produce a temporary set of associations between electronic documents and facts as shown in Table 1, which may then be placed in longer term storage in anticipation searches for those search terms occurring in the future. The search engine then displays the relevant documents or a link thereto along with the relevant facts returned by the fact extraction tool 120 at extraction operation 306 at display operation 308.
FIG. 4 shows the multi-stage approach utilized by embodiments of the fact extraction tool 120. Initially, the fact extraction tool 120 attempts to recognize a set of factual descriptions from the electronic documents of interest at recognition operation 402. Here, the goal is to find those descriptions in the text that are likely to be facts based on finding matches to a fact-word table discussed in more detail below with reference to FIG. 5. By performing a quick matching process, much of the electronic document that should be ignored when finding facts can be eliminated from further fact extraction processing thereby increasing the efficiency of the subsequent stage(s) that are employed to increase accuracy.
After having identified a set of factual descriptions for a document being analyzed, fact extraction is then performed on that set of factual descriptions at extraction operation 404. Here, more detailed analyses are performed only on the set of factual descriptions, as opposed to the whole document, so that satisfactory efficiency is maintained while adequate accuracy is achieved. The analyses of extraction operation involve decision making based on a determination of linguistic constituents of the factual descriptions. Such linguistic constituents may include the syntactic constituents, the semantics, and so forth.
FIG. 5 shows an example of details of the recognition and extraction operations of FIG. 4. The logical operations begin at scanning operation 502 where the fact extraction tool 120 scans the electronic document to find words or phrases matching those of a fact-word table. A fact-word table is a list of words or phrases that are known to likely be used when expressing a fact as opposed to an opinion for example. Table 2 shows a brief example. Note that to provide optimal processing performance, the words of the table may be associated with the most appropriate part of speech (POS) tag which is discussed below in relation to tag operation 504.

	TABLE 2

	Fact-Word List	POS Tags

	Word/Phrase1	POS Tag
	Word/Phrase 2	POS Tag
	Word/Phrase N	POS Tag

Research has been done to determine words that are suggestive of facts rather than opinions. For example, the class of words that introduce facts can be derived using research and work on the classification of verbs and their lexical functions. Two relevant papers that may be used as a material to do so include:

- (1) Mel'cuk (1996) Lexical Functions: A Tool for the Description of Lexical Relations in the Lexicon. In L. Wanner (ed.): Lexical Functions in Lexicography and Natural Language Processing, Amsterdam/Philadelphia: Benjamins, 37-102.
- (2) Fontenelle, T. (1997): “Discovering Significant Lexical Functions in Dictionary Entries”, in Cowie, A P. (ed.) Phraseology: Theory, Analysis, and Applications, Oxford University Press, Oxford.

Thus, on the basis of such research, the fact-word list as shown in Table 2 may be constructed to include those verbs or other words that are suggestive of a fact expression as opposed to a non-fact. For example, the terms “invented” or “hired” are suggestive of a fact expression whereas the terms “can be” or “complains” are not. A particular example of a fact-word list can be found in Appendix A located at the end of this specification. This particular example is a non-exhaustive list of verbs that are fact-words that may be used to discover factual descriptions in electronic documents.
Either upon application of the fact-word table to an electronic document, or in parallel with the application of the fact-word table such as where the POS Tag is already associated with the words in the fact-word table, the parts of speech (POS) of each of the words of each factual description are tagged at tag operation 504. This tagging operation 504, which may occur in parallel with or subsequent to scan operation 502, may involve making disambiguating choices for words which have more than one POS tag, such as by favoring a noun tag over a verb tag since it is understood that syntactic phrases like noun phrases are known to be the entities involved in a factual event. Any unknown and non-pre-tagged words may default to nouns for this reason as well. As with nouns, adjectives may be favored over verbs (e.g., “planned” as an adjective over “planned” as a verb) as well such that words having both an adjective and verb tag will default to adjective because adjective are part of noun phrases which are known to be the entities involved in a factual event. When creating the associations of the POS Tags to the words of the fact-word table, such as when creating the table, these disambiguating choices may already be applied so that, for instance, “planned” is associated with an adjective POS Tag in the table and not a verb POS Tag.
Once the factual descriptions have been found and the words of the factual descriptions have been tagged with the POS, then the more complete analysis may be performed to improve the accuracy of the fact extraction without requiring that the entire document be subjected to this more complete processing. At identification operation 506, syntactic phrases like noun phrases and verb phrases are identified. The syntactic phrases are identified by utilizing conventional grammar rules and light linguistic analysis. Those syntactic phrases that are in the neighborhood, i.e., very local to the set of factual descriptions in a document are identified and if a factual description has no syntactic phrases associated to it, then the corresponding sentence may be eliminated from further consideration. Thus, by focusing on only those syntactic phrases that are in the neighborhood of the factual description, the process avoids looking at all the linguistic constituents of a whole sentence.
Furthermore, at identification operation 506, the linguistic constituents of the factual descriptions having the neighboring syntactic phrases are further determined by assessing the role a syntactic phrase plays within the corresponding sentence based on the pattern identified in the factual description. Thus, it is determined from the word pattern of the factual description whether the syntactic phrase plays the role of subject or object within the sentence containing the current factual description being analyzed.
Once the linguistic constituents of the factual descriptions are determined, i.e., the syntactic phrases and their roles have been identified, exclusion rules may then be applied to those noun phrases of the factual descriptions to further eliminate those that are less likely to be an expression of fact at exclusion operation 508. The exclusion rules may be applicable on the basis of a syntactic phrase as an object, a syntactic phrase as a subject, or a syntactic phrase without regard to its role. Furthermore, in this particular embodiment, an exclusion rule being applied to individual words, to the syntactic phrases, or to the whole sentence lead to the same result, which is to exclude the whole sentence from being a factual description. An example of exclusion rules that may be applied is shown in Table 3.

TABLE 3

Exclusion Rules	Conclusion

“Object” has “opinion/biased” modifier	Rule out the sentence
	candidate
Sentence Filters:	Rule out the sentence
Initial word of sentence (e.g., pronouns)	candidate
Punctuation: e.g. ‘?’
“Subject” is a definite - unless Proper name	Rule out the sentence
	candidate
Surrounding “Context” of the “Object”	Rule out the sentence
	candidate if the
	surrounding context
	has a particular POS
	that is not indicative of
	a fact (e.g., some class
	of pronouns)
Stop words occur in the sentence	Rule out the sentence
	candidate
“Subject” of “Object” contain pronouns	Rule out Noun Phrase

Either upon application of the exclusion rules, or in parallel with the application of the exclusion rules, scoring rules are applied at scoring operation 510. The scoring rules give a weight to both the subject and object noun phrases for each of various features, and a total score for the candidate factual description is the sum of the individual feature weights plus the certainty score of the matching fact-word. The individual feature weights may be positive, when indicative of a fact, and may be negative, when indicative of a non-fact. Examples of features and associated scoring rules are provided below in Table 4. The feature scores may be manually assigned using human judgment or may be automatically learned.

TABLE 4

Features	Scoring rules

Certainty score of the matching pattern
(fact-word, e.g., main verb)
Class of the Roles (i.e., subject or verb),	Score per class
e.g.: person, country, organization, etc.
Main “subject” contains a Proper Name	Normal weight
“Object” length	Length score
“Subject” length	Length score
Sentence length	Length score
“Subject” appears at beginning of sentence -	Position score
i.e., subject offset
“Object” has a modifier (adjective, adverbs)	Negative - Basic
	weight
“Object” is a definite (“the”)	Negative - Basic
	Exclusive when ends
	copula sentence

The total score for the factual description is then compared to a pre-defined threshold to determine whether the total score exceeds the threshold at query operation 512. If the threshold is not exceeded, then the corresponding factual description may be discarded. If the threshold is exceeded, then the factual description, the complete sentence, and/or the complete paragraph or other document portion may be presented as a fact at presentation operation 514. This presentation may include displaying the fact, saving the fact to a library, and so forth.
In utilizing the scoring rules and threshold comparison, the weights assigned to the features and/or the threshold value may be manipulated without manipulating the whole approach to fact extraction. In this manner, the degree of accuracy of fact extraction and presentation can be controlled while the processing steps remain the same.
FIG. 6 shows an example screenshot 600 resulting from performing a search. Search terms have been entered in search field 602 to conduct the search. The search term has been matched to various web site links 604 available from the Internet. The user may visit the electronic documents in the normal fashion.
Additionally, facts 610, 612, and 614 about the search term are displayed in section 608. Accordingly, a user can quickly spot facts about the subject of the search without having to visit any of the electronic documents that have been found and without having to manually read and discern fact from opinion. In this particular example, the facts 610, 612, and 614 include hyperlinks that the user may select to give more information about the source of the fact and/or to show the context within which the facts were discovered (e.g., date of the fact associated, other facts, etc.).
It will be appreciated that screenshot 600 is merely one example of how the facts may be presented to the user. Rather than presenting them in a separate column as shown, they may be listed as sub-elements of the electronic document that they have been extracted from. Furthermore, as an alternative to or in addition to listing the facts on the search results page, the facts extracted from a particular electronic document may also be listed in a column or other location upon the user viewing the electronic document itself. Additionally, as an alternative to or in addition to separating the facts from the document for display, the facts may be highlighted within the electronic documents both in the list of documents 604 within the search results and within the complete electronic document when it is chosen for display. As yet another alternative, the facts may be displayed independently from search results, such as to display facts only with a selectable link to obtain the source documents, where only the extracted facts have been searched to thereby avoid the document search completely.
Additionally, it will be appreciated that the presentation of the extracted facts, such as that shown in screenshot 600, may be provided as a display to a local computer implementing the search and fact extraction for a local user. Alternatively, the presentation of the extracted facts, such as that shown in screenshot 600, may be provided as a display to a remote computer that has requested that the local computer perform the search and fact extraction on its behalf, such as in the case of an Internet based search engine.
Accordingly, facts may be efficiently and accurately extracted from documents for presentation to users. Through the multi-stage approach, the efficiency is increased by avoiding detailed analysis of the entire documents as well as avoiding detailed analysis of the entire sentence where a factual description has been found. The accuracy is maintained by employing further analysis upon the factual descriptions that have been discovered in the document by the initial stage of processing.
While the invention has been particularly shown and described with reference to various embodiments thereof, it will be understood by those skilled in the art that various other changes in the form and details may be made therein without departing from the spirit and scope of the invention. For example, certain exclusion rules that are not specific to the linguistic constituents of a factual description, such as those based on punctuation of a sentence, may be applied when parsing for the factual description rather than later during the application of other exclusion rules.

APPENDIX A

Fact Words

	abase
	abate
	abort
	abrade
	abridge
	absorb
	abstract
	accelerate
	accent
	accept
	accredit
	achieve
	act
	add
	address
	adduce
	adjust
	administer
	admit
	advance
	advertise
	aerate
	afford
	aggravate
	agree
	aid
	aim
	air
	allay
	alleviate
	alter
	amend
	amplify
	amuse
	animate
	announce
	answer
	antedate
	appear
	appease
	apply
	argue
	arouse
	arrange
	arrest
	arrive
	ask
	assemble
	assert
	asseverate
	assign
	assuage
	assure
	attach
	attack
	attenuate
	avert
	avoid
	awake
	award
	back
	bail
	bank
	bar
	barbarize
	bare
	base
	batter
	beach
	beam
	bear
	become
	befog
	befuddle
	beget
	begin
	begrime
	belch
	belie
	bend
	benumb
	bequeath
	bestow
	betray
	better
	bind
	blackleg
	blanket
	bleach
	blemish
	blend
	blight
	blister
	block
	blockade
	blow
	blunder
	blunt
	blur
	blurt
	bob
	bog
	boil
	bolster
	boost
	bowdlerize
	bowl
	brace
	brand
	brave
	break
	brief
	brighten
	bring
	broadcast
	bruise
	buckle
	build
	bull
	bunch
	bundle
	bung
	burlesque
	burn
	burst
	bury
	buy
	bypass
	canvass
	cap
	capitalize
	carry
	cast
	castigate
	castrate
	catch
	chafe
	change
	channel
	charge
	check
	chill
	chime
	chip
	chock
	choke
	choose
	churn
	cipher
	circulate
	circumvent
	claim
	clash
	clean
	cleanse
	clear
	climb
	clinch
	clip
	clog
	close
	clot
	cloud
	cockle
	coin
	collapse
	collect
	colour
	comfort
	commission
	commit
	communicate
	compare
	complete
	compound
	compress
	compromise
	conceal
	concede
	conceive
	conciliate
	conclude
	conduct
	confess
	confide
	confirm
	confound
	confuse
	congeal
	connect
	conserve
	consolidate
	constitute
	constrain
	constrict
	continue
	contort
	contract
	control
	convert
	convey
	cook
	cool
	cordon
	correct
	corrode
	corrupt
	counter
	countersink
	cover
	crack
	crank
	crash
	craze
	create
	cripple
	crop
	cross
	crumble
	crush
	cry
	curb
	curdle
	curtail
	cushion
	cut
	damage
	damp
	dance
	dangle
	darken
	darn
	dash
	deaden
	deal
	debase
	debauch
	debunk
	decay
	decide
	declare
	deepen
	deface
	defeat
	defend
	deflate
	deflect
	deform
	defrost
	delay
	delegate
	deliver
	demise
	demonstrate
	dent
	deny
	deplete
	depreciate
	depress
	deprive
	depute
	derange
	describe
	desecrate
	design
	designate
	desolate
	despoil
	destroy
	detail
	detect
	deteriorate
	determine
	develop
	die
	differentiate
	diffuse
	dilute
	dim
	diminish
	direct
	dirty
	disable
	disappear
	discharge
	discipline
	disclose
	discolour
	disconnect
	discontinue
	discover
	discuss
	disfigure
	disguise
	dislocate
	dislodge
	dismantle
	dismount
	disorder
	dispatch
	dispense
	disperse
	display
	dispute
	disrupt
	distil
	distinguish
	distort
	disturb
	divert
	divide
	dock
	doctor
	dodge
	double
	douse
	draft
	dramatize
	draw
	dredge
	dress
	drive
	drop
	drown
	duff
	dull
	earth
	ease
	eat
	educate
	effect
	elevate
	elicit
	elude
	emancipate
	embellish
	embitter
	embody
	emit
	emphasize
	enable
	encourage
	end
	endorse
	endow
	enforce
	engage
	enhance
	enjoin
	enlarge
	enliven
	ennoble
	enrich
	enrol
	enshrine
	entail
	entangle
	enthrone
	entrust
	enunciate
	epitomize
	equalize
	erect
	escalate
	establish
	evade
	evaporate
	evince
	evoke
	exacerbate
	exact
	exaggerate
	examine
	exasperate
	exceed
	excite
	exhale
	exhibit
	exist
	expand
	expedite
	explain
	expose
	expound
	express
	extend
	extinguish
	extort
	extract
	fabricate
	face
	fade
	fail
	fake
	fall
	falsify
	familiarize
	fasten
	father
	fatten
	feature
	feed
	ferry
	fertilize
	festoon
	fiddle
	fight
	fill
	filter
	finalize
	find
	finish
	fire
	fit
	fix
	flag
	flash
	flaunt
	flay
	float
	flood
	floodlight
	flourish
	flush
	fly
	fog
	foil
	fold
	follow
	force
	forge
	forgive
	form
	foster
	foul
	found
	frame
	fray
	free
	freeze
	frustrate
	furl
	furnish
	furrow
	fuse
	gain
	gallop
	garble
	gash
	generate
	gerrymander
	get
	give
	gladden
	glorify
	gloss
	glut
	go
	govern
	grade
	graduate
	grant
	grate
	graze
	ground
	group
	grow
	guide
	halt
	halve
	hamper
	handle
	happen
	harass
	harbour
	harden
	harm
	harmonize
	harry
	hasten
	hatch
	head
	heal
	hear
	heat
	heighten
	help
	hide
	hit
	hoard
	hoist
	hold
	hope
	hound
	hurt
	identify
	illuminate
	imagine
	impair
	impart
	impeach
	impede
	imperil
	implant
	improve
	inaugurate
	increase
	indent
	indenture
	indicate
	induce
	induct
	infect
	infiltrate
	infix
	inflame
	inflate
	inflict
	influence
	inform
	infuse
	initial
	initiate
	injure
	insert
	inspire
	instigate
	instil
	institute
	integrate
	intend
	intensify
	interpolate
	interrupt
	intimate
	introduce
	invert
	invigorate
	invite
	invoke
	involve
	issue
	jab
	jam
	jettison
	jingle
	join
	jumble
	jump
	justify
	keep
	kick
	kill
	kindle
	knock
	lacerate
	ladder
	lance
	land
	laugh
	launch
	lay
	layer
	lead
	leave
	lend
	lengthen
	lessen
	let
	level
	liberate
	lie
	light
	lighten
	limit
	line
	link
	listen
	litter
	live
	liven
	load
	lock
	loose
	loosen
	lose
	lower
	lump
	magnify
	maintain
	make
	manage
	mangle
	manipulate
	manufacture
	mark
	marshal
	mask
	match
	matter
	maul
	measure
	meet
	mellow
	melt
	mend
	mention
	mildew
	mind
	misrepresent
	miss
	mist
	mitigate
	modify
	mollify
	moot
	mould
	move
	muddle
	muddy
	muffle
	muss
	muster
	mute
	mutilate
	narrow
	navigate
	neaten
	nick
	nip
	notch
	notice
	nourish
	nurse
	obfuscate
	obscure
	obstruct
	obtain
	occupy
	occur
	offend
	offer
	open
	operate
	oppose
	order
	originate
	outline
	overcharge
	overdo
	overflow
	overturn
	overwork
	pacify
	pack
	pad
	panic
	paralyze
	pare
	parlay
	parole
	parry
	part
	partition
	pass
	patch
	pay
	peal
	peddle
	peg
	penalize
	perform
	perish
	persecute
	pervert
	phrase
	pick
	pillow
	pique
	pit
	placard
	place
	plan
	plant
	play
	pluck
	plug
	plunge
	point
	poison
	pole
	polish
	poll
	pool
	pop
	pose
	position
	post
	pound
	preach
	precipitate
	predate
	prefer
	prejudice
	preoccupy
	prepare
	present
	preserve
	prettify
	prevent
	prick
	prime
	proclaim
	procure
	produce
	profess
	programme
	promote
	promulgate
	prop
	propagandize
	propel
	propound
	prosecute
	protect
	protest
	prove
	provide
	provoke
	prune
	publicize
	publish
	pull
	pulp
	punch
	puncture
	punish
	punt
	purge
	push
	put
	qualify
	quarter
	quench
	question
	quicken
	quieten
	quilt
	race
	raise
	ransack
	rap
	rationalize
	rattle
	re-engage
	re-establish
	re-form
	read
	rear
	reawaken
	recall
	receive
	reclaim
	recline
	recognize
	recommend
	reconcile
	reconsider
	record
	recruit
	reduce
	refer
	refine
	reflect
	refloat
	reform
	refuse
	regard
	register
	regulate
	rehabilitate
	rehearse
	reinforce
	reissue
	reject
	rekindle
	relate
	relax
	release
	relieve
	reline
	remould
	remove
	rend
	renew
	renovate
	reopen
	repair
	replace
	report
	republish
	require
	rerun
	reseat
	resist
	rest
	restart
	restore
	restrain
	result
	resurrect
	retail
	retain
	retire
	retract
	retrench
	retrieve
	return
	reveal
	reverse
	revive
	rewind
	right
	ring
	rise
	roast
	rock
	roll
	rotate
	rouse
	row
	ruffle
	ruin
	rumple
	run
	rush
	rustle
	sail
	salvage
	sap
	save
	scald
	scorch
	score
	scotch
	scratch
	scream
	scuff
	scupper
	scuttle
	seal
	sear
	seat
	secure
	see
	sell
	send
	serve
	set
	settle
	sever
	shake
	shame
	sharpen
	shatter
	sheathe
	shed
	shelter
	shield
	shift
	shine
	shingle
	shirk
	shoot
	shorten
	shout
	show
	shrink
	shut
	sift
	sign
	signal
	signalize
	signify
	simmer
	sing
	singe
	sink
	sit
	site
	situate
	skirt
	slacken
	slake
	slash
	sleep
	slice
	slip
	slow
	smear
	smile
	smudge
	snag
	snap
	snarl
	snuff
	sober
	soften
	soil
	solace
	solidify
	soothe
	sort
	sound
	sour
	sow
	spare
	spark
	speak
	speck
	speed
	spill
	spin
	splinter
	split
	splodge
	spoil
	sponsor
	sport
	spot
	spout
	sprain
	spray
	spread
	spring
	square
	squash
	squeeze
	stack
	staff
	stain
	stalemate
	stall
	stamp
	stand
	star
	starch
	start
	staunch
	stay
	steady
	steer
	stem
	step
	stick
	stiffen
	still
	stir
	stoke
	stop
	store
	straighten
	strain
	strand
	strengthen
	stress
	stretch
	strike
	strip
	strum
	study
	stuff
	stultify
	stunt
	subdue
	subscribe
	subvert
	succeed
	suffer
	suggest
	suit
	summarize
	supplement
	supply
	support
	suppose
	suppress
	surface
	surrender
	survive
	suspend
	sustain
	sweep
	sweeten
	swell
	swing
	swish
	taint
	tarnish
	task
	teach
	tear
	telephone
	temper
	tend
	thank
	thaw
	thin
	thrill
	throw
	thrust
	thump
	thwart
	tidy
	tighten
	toll
	tootle
	topple
	torment
	torture
	total
	touch
	toughen
	tousle
	tow
	train
	trample
	transfer
	transplant
	trap
	travel
	treat
	trigger
	trim
	truss
	try
	tumble
	turn
	twang
	twiddle
	twirl
	twist
	unblock
	unburden
	unclog
	undo
	unfasten
	unfix
	unfold
	unhinge
	unhitch
	unite
	unloose
	unravel
	unsaddle
	unseat
	unsex
	unstop
	untangle
	untwist
	uphold
	upset
	urge
	use
	validate
	vandalize
	veer
	veil
	ventilate
	vocalize
	voice
	vote
	vulgarize
	waft
	waggle
	wake
	walk
	wangle
	warm
	warn
	warp
	warrant
	wash
	watch
	weaken
	wean
	wear
	weave
	weep
	weld
	whet
	whirl
	whitewash
	widen
	wield
	wiggle
	wilt
	win
	wind
	wing
	wipe
	wire
	wish
	withdraw
	wither
	withhold
	work
	worry
	wreak
	wreck
	wrest
	wring
	wrinkle
	write
	yield

Claims

1. A method of finding facts within electronic resources, comprising:

scanning an electronic resource to discover factual descriptions of sentences that comprise words matching words of a fact-word table;

examining the discovered factual descriptions to identify the linguistic constituents of the factual descriptions; and

determining whether to present a factual description as a fact based on the identified linguistic constituents.

2. The method of claim 1, wherein determining whether to present a factual description as fact based on the identified linguistic constituent comprises:

applying excluding rules in relation to the linguistic constituents of the factual descriptions to eliminate certain factual descriptions from consideration;

scoring the factual descriptions;

comparing the score of each factual description remaining for consideration to a threshold; and

for each factual description having a score that exceeds the threshold, presenting at least a portion of the sentence containing the factual description as a fact.

3. The method of claim 2, further comprising tagging words of the factual descriptions with their parts of speech.

4. The method of claim 3, wherein tagging words of the factual descriptions with their parts of speech comprises applying a noun tag when a word may be either a verb or a noun.

5. The method of claim 4, wherein applying the excluding rules comprises applying a first set of rules for syntactic phrases that have a role of subjects and applying a second set of rules for syntactic phrases that have a role of objects.

6. The method of claim 5, wherein applying the first set of rules comprises excluding noun phrases having an opinion or biased modifier of subjects or objects.

7. The method of claim 5, wherein applying the second set of rules comprises excluding subject noun phrases which non-proper name definite descriptions, excluding noun phrases which conatain pronouns, and excluding subject noun phrases which do not appear at the beginning of text.

8. The method of claim 5, further comprising applying a third set of rules without regard to the role of the noun phrase.

9. The method of claim 8, wherein applying the third set of rules comprises excluding factual descriptions where the punctuation of the sentence is a question mark, and excluding sentences with phrases that include a stop word.

10. The method of claim 2, wherein scoring the factual descriptions comprises scoring only those factual descriptions remaining for consideration either after or during application of the excluding rules.

11. A computer readable medium containing instructions that perform acts comprising:

receiving a search term;

parsing a plurality of electronic documents to discover factual descriptions of sentences that comprise words matching words of a fact-word table;

determining whether to present a factual description as a fact relevant to the search term based on the identified linguistic constituent.

12. The computer readable medium of claim 11, wherein the acts further comprise obtaining the plurality of documents by searching an collection of electronic documents to find those documents containing the search term, wherein the collection is searched to find those documents containing the search term prior to parsing the plurality of electronic documents.

13. The computer readable medium of claim 11, wherein the acts further comprise obtaining the electronic documents and presenting factual descriptions prior to receiving the search term and searching the electronic documents and factual descriptions to find those electronic documents and corresponding factual descriptions that are relevant to the search term.

14. The computer readable medium of claim 11, wherein determining whether to present a factual description as a fact relevant to the search term based on the identified linguistic constituent comprises:

applying excluding rules in relation to the linguistic constituents of the factual descriptions to eliminate a portion of the factual descriptions from consideration;

scoring the factual descriptions;

for each factual description that is taken from an electronic document that contains the search term and that has a score that exceeds the threshold, presenting at least a portion of the sentence containing the factual description as a fact relevant to the search term.

15. The computer readable medium of claim 14, wherein scoring the factual descriptions comprises scoring only those factual descriptions remaining for consideration after applying the excluding rules.

16. A computer system, comprising:

storage containing a plurality of electronic resources that comprise textual information;

a processor that receives a request to present facts that are related to the search term from a set of electronic documents, wherein the processor parses the plurality of electronic documents to discover factual descriptions of sentences that comprise words matching words of a fact-word table, examines the discovered factual descriptions to identify the linguistic constituents of the factual descriptions, determines whether to present a factual description as a fact based on the identified linguistic constituent, and presents at least a portion of sentences that contain the factual descriptions that are determined to be presented as a fact and that are related to the search term.

17. The computer system of claim 16, further comprising a display device and wherein the processor presents at least the portion of the sentences by displaying at least the portions of the sentences on the display device.

18. The computer system of claim 16, further comprising a network interface and wherein the processor presents at least the portion of the sentences by outputting those portions to another computer via the network interface.

19. The computer system of claim 16, further comprising a network interface and wherein the storage is accessible by the processor via the network interface.

20. The computer system of claim 16, wherein the processor determines whether to present a factual description as fact by:

scoring the factual descriptions;

for each factual description that contains the search term and that has a score that exceeds the threshold, presenting at least the portion of the sentence containing the factual description as a fact relevant to the search term.